Sensitivity analysis of EKF and iterated EKF pose estimation for position-based visual servoing

Abstract

Robust and real-time relative pose estimation is an integral part of a position-based visual servoing (PBVS) system. Traditionally, extended Kalman filter (EKF) has been used to solve for the nonlinear relative end-effector to object pose equations from a set of 2D-3D point correspondences. However, the performance of the estimation filter and the convergence of the pose estimates are highly sensitive to tuning of filter parameters, camera calibration, and image processing. In this paper, the application of iterated EKF (IEKF) for a robust high-speed PBVS system is studied. We also provide a detailed analysis of the stability and sensitivity of the EKF and IEKF pose estimation to uncertainties in (1) tuning of filter parameters, namely, process and measurement noise covariance matrices, initial state estimate, and sampling time (speed of PBVS system), (2) features selection, and (3) calibration of camera intrinsic parameters. Experimental results show that IEKF outperforms the standard EKF without bandwidth sacrifice and should be used to improve the robustness of the PBVS system to uncertainties